Wenduo Wang
Wenduo Wang
@dalcinl Could you please rebase the PR? Hats off to @hppritcha for chasing down the problem(s).
Yes you are right. We are doing it in https://github.com/open-mpi/ompi/pull/12591
You have helped a lot. I will rebase #12591 after merging this PR.
@mjwilkins18 Curious if you could try out #12552 and see if that helps?
@bosilca Could you please resolve the conflict?
@jiaxiyan Please fix the build.
@bosilca Since this PR is backward compatible and does not cause performance regression for the default k=2 case, I wonder if we can merge it and work on improvement as...
@bosilca Makes sense. Jessie has rotated to another high-priority project. I will take a look at the latency jump.
@bosilca I did an analysis of the algorithm with EFA. The weird performance bumps at some message sizes is a result of 2 factors: - The number of send-recv rounds...
@edgargabriel Could you explain how this change can be tested? We can run it through AWS CI.